What Is Predictive Maintenance and How Does It Work?

Every building system degrades over time — but the rate at which it degrades is not fixed. A pump bearing that runs hot, a fan motor with increasing vibration, a boiler flue with drifting combustion efficiency: each of these conditions develops gradually, produces measurable signals, and can be addressed before it causes a failure. Predictive maintenance is the discipline of reading those signals systematically and acting on them at the optimal moment — after condition has deteriorated enough to justify intervention, but before failure occurs. In this article we explain what predictive maintenance is, how it differs from preventive and reactive approaches, which methods are used, and what building operators need to implement it effectively.

Predictive maintenance (PdM) is a condition-based maintenance strategy in which the actual physical state of equipment is monitored continuously or at defined intervals, and maintenance work is triggered when that state indicates that intervention will be needed within a defined timeframe. The goal is to perform maintenance at the right time — not too early (wasting resources on equipment that still has useful life remaining) and not too late (after failure has already caused downtime or damage).

This distinguishes predictive maintenance from the two other main approaches. Reactive maintenance waits for equipment to fail before acting — minimising upfront cost but accepting unpredictable downtime, high emergency repair costs, and the knock-on consequences of sudden failures in interconnected building systems. Preventive maintenance replaces or services components on a fixed schedule regardless of their actual condition — more reliable than reactive, but often wasteful when components are serviced or replaced while still in good condition.

Predictive maintenance sits above both: it uses data about actual equipment condition to time interventions precisely. The trade-off is that it requires investment in monitoring technology, data collection infrastructure, and the analytical capability to interpret what the data shows. For building portfolios where equipment downtime is costly, compliance obligations are strict, or energy efficiency targets demand continuous optimisation, that investment is increasingly justified by the savings it generates.

Predictive maintenance relies on techniques that reveal the current condition of equipment without taking it out of service. The table below shows the main methods used in building maintenance, what each measures, and where it is typically applied.


Note: in practice, most predictive maintenance programmes combine multiple methods. IoT sensor monitoring provides continuous low-cost baseline data; specialist techniques such as vibration analysis and thermography are applied periodically or triggered when sensor data indicates a developing anomaly.

Understanding where predictive maintenance fits in the broader maintenance strategy requires a clear comparison of all three approaches. Each has legitimate applications — the question is which is appropriate for which assets and contexts.

• Reactive maintenance (run-to-failure): lowest upfront cost, no monitoring investment required. Appropriate for non-critical assets where failure has no significant safety, compliance, or operational consequence — a broken light fitting, a failed door closer. Entirely inappropriate for safety-critical systems, energy-intensive equipment, or assets whose failure cascades into wider building system disruption.
• Preventive maintenance (time-based): scheduled at fixed intervals regardless of equipment condition. Reliable and auditable — the FM team knows exactly when work was last done and when it is next due. The limitation is efficiency: a bearing replaced at 12 months because the schedule says so may have had 18 months of useful life remaining. Appropriate for equipment where condition monitoring is not cost-effective, or where regulatory requirements mandate fixed inspection intervals regardless of condition.
• Predictive maintenance (condition-based): triggered by actual equipment condition rather than a fixed schedule. Maximises component life, reduces unnecessary interventions, and provides early warning of developing failures. Requires investment in monitoring technology and analytical capability. Most cost-effective for high-value assets, energy-intensive equipment, or systems where failure has significant consequences.
• Combined approach: most mature FM programmes use all three in parallel, assigning the appropriate strategy to each asset class based on criticality, failure consequence, monitoring cost, and regulatory requirements. Predictive maintenance is applied to the highest-value and highest-risk assets; preventive to mid-range assets with defined regulatory cycles; reactive to low-criticality items

Condition-based maintenance (CBM) — in German: zustandsbasierte Instandhaltung — is the underlying principle on which predictive maintenance is built. CBM holds that maintenance should be performed when equipment condition indicates it is needed, not when a calendar date arrives. Predictive maintenance extends CBM by using advanced monitoring technologies and data analysis to detect condition changes earlier and with greater precision than traditional inspection methods.

The practical implementation of CBM requires defining, for each monitored asset, the condition parameters that will be measured (temperature, vibration, current draw, pressure differential), the baseline values for normal operation, the threshold values that indicate developing degradation and trigger a maintenance alert, and the critical values that indicate imminent failure and require immediate intervention.

These thresholds are typically derived from manufacturer specifications, historical failure data from similar equipment, and the professional judgement of maintenance engineers. Over time, as monitoring data accumulates, thresholds can be refined based on actual observed failure patterns for specific equipment in specific operating environments. This is where the long-term value of a predictive maintenance programme becomes clear: the longer it runs, the more precisely it predicts failure timing, and the more efficiently maintenance resources are deployed.

Not every building system is an equally good candidate for predictive maintenance. The best candidates are those where condition degradation produces measurable signals before failure, where failure has significant operational or financial consequences, and where the cost of monitoring is proportionate to the value of the asset or the cost of failure.

• HVAC and ventilation: rotating equipment (fans, pumps, compressors) produces vibration signatures that change predictably as bearings wear. Heat exchanger performance can be tracked through differential temperature measurements. Coil fouling shows as increased fan current draw and reduced airflow. IoT sensors on HVAC systems are among the most cost-effective predictive maintenance investments in commercial buildings.
• Lifts and escalators: lift machinery is a natural candidate for vibration and current monitoring. Motor current signatures reveal developing winding faults before they cause shutdown. Door mechanism wear shows in current draw anomalies. Given the safety-critical nature of lift equipment and the cost of emergency call-outs, predictive monitoring typically pays back quickly.
• Electrical systems: power quality monitoring identifies developing transformer and switchgear problems through harmonic distortion and voltage anomaly patterns. Thermal imaging of distribution boards detects loose connections and overloaded circuits before they cause fires or failures. Particularly important in facilities with high electrical load density or sensitive equipment.
• Pumping systems: pressure and flow monitoring tracks pump performance against design curve. Efficiency degradation indicates impeller wear or seal deterioration. Vibration analysis detects cavitation and bearing wear. Booster sets, chilled water pumps, and heating circulation pumps all benefit from condition monitoring.
• Building fabric and roofing: thermal imaging surveys detect moisture ingress in flat roofs before it reaches the internal structure. Crack monitoring sensors on structural elements track movement over time. While less amenable to continuous monitoring than mechanical systems, periodic condition assessment of building fabric significantly reduces the risk of undetected degradation becoming major repair work.

Implementing a predictive maintenance programme requires four components working together: monitoring hardware, data infrastructure, analytical tools, and qualified people to act on the outputs.

Monitoring hardware ranges from simple IoT sensors — temperature, humidity, current clamps, pressure transducers — to specialist instruments such as vibration analysers and thermal cameras. The cost of IoT sensors has fallen sharply over the past decade, making continuous monitoring of a wide range of building systems economically viable for portfolios that previously could not justify the investment. Specialist instruments are typically provided by certified contractors rather than owned by the FM team.

Data infrastructure means the connectivity and storage systems that carry sensor data from equipment to the platform where it is analysed. Most modern building management systems can integrate IoT sensor data; standalone platforms also exist. The key requirement is reliable, secure connectivity from equipment to data store, with appropriate data retention periods for trend analysis.

Analytical tools range from simple threshold alerting — the system sends a notification when a parameter exceeds a defined value — to machine learning models that identify anomalous patterns across multiple data streams and predict failure timing with quantified confidence. Simpler approaches are sufficient for many building maintenance applications; advanced analytics deliver the greatest value for high-complexity equipment or large portfolios where manual data review is not practical.

Qualified people are the final and most important component. Sensor data and analytical outputs must be interpreted by maintenance engineers who understand the equipment, can distinguish meaningful signals from background noise, and can translate data insights into appropriate maintenance actions. Predictive maintenance does not replace skilled technicians — it gives them better information on which to base their decisions.

Predictive maintenance programmes fail or stall for predictable reasons. The following barriers appear consistently across organisations attempting to implement PdM for the first time:

• Upfront investment hesitation: the cost of sensors, connectivity, and analytical platforms is visible immediately; the savings from avoided failures are distributed over time and harder to attribute. The business case requires realistic modelling of avoided downtime costs, reduced emergency repair spend, and extended asset life — not just a comparison of monitoring costs against scheduled maintenance costs.
• Data quality and integration problems: sensors that drift, connectivity that drops, and data stored in formats that cannot be integrated with existing maintenance management systems all undermine the programme’s reliability. Piloting with a small number of high-priority assets before portfolio-wide rollout allows data quality issues to be resolved at manageable scale.
• Skills gap in data interpretation: an alert from a vibration monitoring system is only useful if the maintenance team knows what it means and what to do about it. Training FM staff to interpret condition monitoring outputs, or partnering with specialist contractors who provide monitoring-plus-response services, addresses this directly.
• Organisational resistance to changing maintenance schedules: time-based preventive maintenance is comfortable because it is predictable and auditable. Moving to condition-based scheduling requires confidence that the monitoring data is reliable, and that deviating from fixed schedules will not create compliance problems. Starting with assets where regulatory requirements do not mandate fixed intervals makes the transition easier.
• Lack of historical failure data: predictive models are more accurate when trained on historical failure data from similar equipment. Organisations implementing PdM for the first time often lack this data. The solution is to start monitoring immediately and build the dataset over time, while using manufacturer specifications and industry benchmarks as initial threshold references.

Predictive maintenance generates alerts; those alerts must be acted on by qualified technicians. For FM teams managing multi-site portfolios, the challenge is not identifying that a pump bearing is showing early signs of wear — it is finding a qualified contractor who can inspect and service that bearing at the right location within the right timeframe, before the condition deteriorates further.

Wowworks connects facility managers with vetted maintenance contractors across all relevant trade disciplines — HVAC engineers, electrical technicians, lift specialists, and building fabric contractors — who can respond to condition-based maintenance alerts as they arise. Through the platform, tasks triggered by predictive monitoring data are assigned digitally, completed work is documented, and the maintenance record is updated in a format that supports both asset management and compliance audit requirements.

As predictive maintenance becomes standard practice in professional facility management — driven by falling sensor costs, tighter energy performance requirements, and the growing availability of integrated building management platforms — the ability to translate monitoring data into timely, qualified maintenance interventions becomes the critical operational capability. Wowworks provides the contractor network that makes that translation reliable at portfolio scale.